ISHS


Acta
Horticulturae
Home


Login
Logout
Status


Help

ISHS Home

ISHS Contact

Consultation
statistics
index


Search
 
ISHS Acta Horticulturae 464: International Postharvest Science Conference Postharvest 96

LINKING QUALITY CHARACTERISTICS WITH EXPERT KNOWLEDGE: OBJECTIVE QUALITY GRADING OF CHICORY BASED ON SIZE, SHAPE AND COLOUR

Authors:   R. De Busscher, I. Paulus, E. Schrevens
Abstract:
Decisions about quality of horticultural products, are mainly based on a subjective, visual interpretation of external characteristics like size, shape, color and color distribution, texture and structure of the surface, defects, contaminations, etc. All the essential information to quantify these characteristics are embedded in one or more digitized video images of the product. Feature extraction from such images leads to a methodology for objective quality characterization. As a model object, chicory heads are used in this study.

It is possible to link the set of quality descriptors extracted out of the images and the expert decision of the product by means of Tree Based Modeling techniques. In the experiment we determined the set of quality descriptors and asked 5 grading experts to rate the quality class of each chicory head. This classification was carried out twice on the same heads without their knowledge so that we could assess the “within expert” misclassification error. The comparison of the two quality decision gradings on each head revealed that in 1/3 of the cases the graders assigned a different quality/auction class. Furthermore there was about the same amount of disagreement between the different persons.

Using Tree Based Modeling, we are able to generate a binary decision tree that simulates the expert grading based on the set of quality descriptors. All these simulations have an error of 35–40% based on an independent test sample. This is about the same magnitude as the repeatability of expert graders. Looking at the models for each person, the difference in importance of distinct quality descriptors becomes obvious.

The study clearly demonstrates the usefulness of Tree Based Modeling in simulating the expert knowledge of horticultural products. On the other hand, some clarifications in the quality descriptions are urgently required resulting in better and easier-to-replicate partitions.

Download Adobe Acrobat Reader (free software to read PDF files)

464_88     464     464_90

URL www.actahort.org      Hosted by K.U.Leuven      © ISHS